Pr[Cartel in Norway]

Size: px
Start display at page:

Download "Pr[Cartel in Norway]"

Transcription

1 F R O N T M A T T E R Pr[Cartel in Norway] Gorm Andreas Grønnevetã«gorm.gronnevet@nhh.no Norwegian School of Economics and Business Administration ABSTRACT. We use data from Norwegian horizontal cartels to estimate the probability of starting a cartel and the average duration of the cartels. We relate this probability to the number of firms in the cartel, in order to investigate the hypothesis that fewer firms make cartel creation more likely and cartel duration longer. We find that that the number of firms have a great impact on probability of cartel formation and the duration of cartels. This suggests that market characteristics are important when predicting the probability of an illegal cartel occurring. JEL: L13, L41, K21, K42, C41 Keywords: Cartel duration, cartel formation, Norway Date: March 11, 2005 Competing for the Young Economist Award: Yes.

2 B O D Y 2 Pr[Cartels in Norway] 1. Introduction This paper investigates cartels in the sense of collusive behavior where the members actually meet to fix prices or share markets. Since cartels are illegal, it seems unlikely that firms would meet illegally to collude if it could be done without meeting. It seems fair, then, to assume that the meetings are necessary. By consequence, our cartels are not a subset of the tacitly collusive regimes that are commonly studied in industrial organization, because in tacit collusion meetings are not necessary. There is no theoretical models that explains why explicit meetings would be necessary in some markets while not in others. The approach taken here is to look at the cartels that were prosecuted and try to infer how their differences may explain the time it took to start the cartel and the duration of the cartel. In fact, we will use the number of firms as the chief explanatory variable and we will find that the number of firms seems to have a great impact on the facility with which the firms collude and the duration of the cartels. A key assumption in our model is that all firms are equal when it comes to the propensity to collude. This is of course a gross simplification. If one believes that the management of some firms are more likely to form cartels than others, one would perhaps feel more comfortable in a model where the probability of continuing in a cartel would be greater than the probability of forming a cartel. But the management changes over time. Suppose for example that the average stay in a position for a CEO is 10 years. 1 If the markets has five firms, the average overlap of a set of managers will be two years. Given that we do not have information regarding the management of each firm, it seems quite fair in this respect to assume a constant (average) probability. As mentioned, we will use the number of firms in the market. On the other hand we are going to ignore other potential important variables such as sunk cost and demand elasticity. For a given competitive outcome it is clear that an elastic demand will make the profits for that cartel smaller than for an inelastic demand. The demand elasticity should be included in the model if one believes that higher returns may help overcome the personal fears of the management. On the same note, it seems likely that high sunk costs will make entry in a market less likely and therefore the expected returns higher. While we would like to take these factors into consideration, we do not at present have an estimate of these variables in our data set.

3 Gorm Grønnevet 3 2. Related literature 2.1. Theoretical models. There are several strands of theoretical models for cartels. [2] proposes a model firms have asymmetric information and where price wars may be used as a mechanism to ensure collusion. Other work on supergames and collusion include [3] and [4]. [5] proposes a dynamic model with finite states and Markov perfect equilibrium. Common for the theoretical literature is that collusion defection will not occur and cartel meetings are not necessary. The theoretical literature is seems to ignore the illegal cartels Empirical work. The empirical work on collusion has been on cartels, as defined in legal literature. Most papers, such as [6] and [7], single out a cartel and then study the dynamics of this cartel. A notable exception to this is [8] who work on a panel of cartels in the US. Their work is also the closest to our. They estimates the probability on being caught, conditional on being caught eventually. While this may look like a tautology, it is not. Their data seem to suggest that the probability of being caught is independent of the duration of the cartel. Under this assumption and the assumption of steady state, the age profile of the cartels when caught will identify the probability of being caught. Of course, their findings are limited to the subsample of the cartels that are eventually caught. If these cartels are more inept than the undetected cartels, their conclusions will not generalize. Their finding are: è The average cartel lasts between 5 and 7 years è Approximately 7 cartels are formed each year è The probability of getting caught in a given year is between 13 % and 17 % è There are at least 36 to 50 active cartels at any point in time (that will be caught) The authors do not pretend to say anything about the cartels that are not caught. The above estimates are therefore conditional upon being caught at one point or another. To illustrate this, we can (with some simplifications) calculate that 7 cartels formed each year and lives for 5 to 7 years will give us 35 to 42 cartels alive at any point in time. Likewise, an average lifetime of 7 years with an equal probability of being caught in any year gives us, again with simplifications, 1/7 = 14 % probability of being caught. As these back of the envelope estimates perhaps illustrates, these estimates are conditional on being caught. This certainly limits their applicability in terms of policy questions.

4 4 Pr[Cartels in Norway] 3. Data 3.1. The Norwegian Legal System. Cartels were prohibited by a government regulation under the Price Act in The prime concern of the authorities at the time was inflation. A series of exemptions to the regulation was therefore given to firms in the two following decades. It is not very easy to pinpoint when the case law under the act changed. But it seems to have changed gradually in the 1980's. In 1988, merger control was introduced and the first illegal cartel was prosecuted the same year The Data. Our data are based on the prosecutions of cartels in the courts. Due to the lenient exemptions for a long while, there are only 26 horizontal cartels in our sample. 2 These cartels have an average duration of 4 years. A picture of the duration is shown below, where the vertical axis is an indicator variable for the cartel and the horizontal axis shows the start date and duration of the cartels. This picture illustrates why there is a truncation on the top of our graph. As the picture shows there are some cartels that are active for a long time before they are discovered. There may for example be an active cartel in 2003 which will be discovered in Figure 1: Duration of cartels sorted by starting date The earliest cartels were recorded by the Court as formed in In some cases this is not really the date the cartes were started, but rather the first time the cartel was illegal. As explained above, the period from 1960 and until the 1980's, there was quite a lenient exemption policy at the Price Authority then in charge of enforcing the antitrust clauses. It is not clear whether 1982 was a turning point in the practice, but the figure above shows quite clearly that there are three cartels that started their illegal activity that the beginning of the following year.

5 Gorm Grønnevet 5 If we sort the data on the date of discovery instead of inception, we get at different picture. Although it is not clear that there is a statistical difference, it seems that the average cartel length has increased in the later years of our sample Figure 2: Duration of cartels sorted by end date The duration of the cartels is one matter. Another is the number of firms involved. The following graph shows a density plot of the number of firms involved in our cartels. As can be seen from this figure, there is one cartel with a large number of firms (47). This was a horizontal price fixing cartel that was organized around a system for calculating prices. Density var1 Figure 3: Density plot of the number of firms in the cartels

6 6 Pr[Cartels in Norway] The main weight is however around about 5 firms This observation is certainly in line with the common wisdom that many firms make cartels more difficult to create and (perhaps) more unstable over time. 4. The estimation method and results 4.1. Preliminary discussion. Our approach to the estimation will be somewhat different from [8]. They assumed that there was cartels were formed as a stochastic process independent of the number of firms involved. Since it seem likely that the number of firms is important for the cartel formation and duration, we would like to take this into account. This means that we move from a descriptive model, to a model the process of cartel formation. Figure 4 below plots the number of firms on the vertical axis and the duration of the cartels on the horizontal axis. There are several short cartels with few members, but all the long cartels have less than seven members. If the probability of getting caught is independent of duration and the number of firms, this picture may suggest that a greater number of firms decreases cartel duration Figure 4: Number of firms (y-axis) versus duration (x-axis) The question is how one should think of the connection between the number of firms and cartel duration. One way to make a link between the number of firms and cartel duration is to assume that there is an individual cost to colluding illegally and that every time the cartel meets, its participants draw a random benefit from continuing. This means that every time the cartel needs to meet, there is a possibility that the cartel may break down. Although it is difficult know why cartels need to meet, there is some evidence that it may be important in terms of controlling the supply and prices for every competitor. This is for example what ADM suggested to the competitors when the lysine cartel seemed to be less effective (see for example [9]). This means that every time the cartel meets, the probability of continuing is the product of each participants probability of agreeing to continue. If every participant has a

7 Gorm Grønnevet 7 probability p of agreeing to the cartel meeting, a cartel of three firms will then continue with probability p 3. We will assume that each firm has to agree to form a cartel in each period that the cartel lasts. This is a binary decision, which suggest that we use a Bernoulli distribution to model the decision. This distribution is assumed to be common to all firms, and the probability for x firms agreeing a cartel will therefore a negative binomial distribution with being caught is the success. 3 The probability of there being a cartel is rather simpler since it is the joint probability that all the firms says yes (p m ). We will use maximum likelihood to estimate the model Assumptions. At present we will proceed to estimate the probability of cartel formation and the duration of the cartels separately. Furthermore, we are going to use two different specifications. The first is a negative binomial model which is independent of the number of firms in the cartel. The negative binomial distribution where the number of firms is ignored is specified as follows: (4.1) J x+r-1 r-1 N J1 - pnx p r where the variable r is the number of successes, which will be taken to be 1. When we take account to the number of firms (m), the distribution looks like this (4.2) K x + r - 1 r - 1 OH1 - pm L x p mr when we model the distribution of cartel formation. As mentioned above, it makes sense to assume that all firms have to agree to the decision for the cartel to be formed, and thus giving us the probability p m. When we model the cartel duration, this does not make sense because it implies that markets with more firms will last longer. In line with the model for cartel formation, we shall therefore model this as 1 - p m. As mentioned above, the distribution will be truncated. This means that cartels of long duration may not have been discovered yet, implying that estimates that do not take this into account may underestimate the probability of cartel formation. Likewise, estimates of cartel duration will be two low if the truncation is not taken into account Results. The estimated probability of cartel formation when we do not account for the number of firms is 0.10 per year. It is slightly harder to report the results for the probability of cartel formation when we account for the number of firms. The average number of participating firms in our sample is seven. For the market of seven firms, we find that the probability of forming a cartel is 0.04 per year. We have to reduce the number of firms in the market to five to find a cartel formation probability of about 0.1 per year. In expectation this means that a market of seven firms will have to wait 24 year before they agree, while a market of five firms only will have to wait 9 years.

8 8 Pr[Cartels in Norway] When we look at the estimate of being discovered, we find that the untruncated estimate is about 0.23 per year and if the truncation is taken into account, the estimate falls to The cartel duration increases, but not by much. This estimate is slightly higher than reported in [8], but probably not significantly different (in a statistical sense). When we look at the number of firms we find that a market of seven firms will a probability of survival of 0.26 while a market of five firms will have a probability of survival of This correspond to expected durations of respectively 2.8 and 4.3 years. Although the final probabilities are based on the data, it must be admitted that the structure imposed on the relationship between the probabilities and the number of firms is probably the main driver of magnitude. Further work will therefore be necessary to find a more flexible way of including the number of firms in the probability model. 5. Conclusion We have found that the probability of forming a cartel and the duration of cartels seems to be in line with estimates in the literature based on US firms. These estimates seems to high to be applicable to any market. We have therefore attempted to make the estimates more realistic by taking into account the structure of the markets through the number of firms. However, work remains. First we would like to make a joint estimation of the probability of forming a cartel and the probability of getting caught. Second, we would like to use the data on the distribution of firms in the Norwegian markets to estimate the total number active of cartels. Third, we will try to make use a more flexible structure to assess the impact of firms on the cartels. Last, we will try to get more information of the antitrust activity in Norway in the 1970 s and 1980 s to get a better understanding of how the Competition Authority worked at that time.

9 B A C K M A T T E R Gorm Grønnevet 9 Notes 1 The average exit rate in Norway is estimated to around 11% - 15 % per year for persons in superior positions. This implies that the average tenure for a senior representative in a firm is between 7 and 10 years. (See [1] ) 2 Two of the cartels were found not guilty in Court, but were included because it was more likely than not that the cartel actually existed. In the first case the Court found not guilty on a legal technicality. The Court found that the procurements in question could not be viewed as an auction and therefore the firms were not guilty of rigging the bids as the prosecutor claimed. They may have been found guilty of price-fixing, but the prosecutor had left this out from the indictment. The case was appealed but the appeal dismissed. In the second case, the Competition Authority had talked to the main government witness (which is illegal in Norway) and therefore totally undermining the credibility of that witness. This case was not appealed. 3 If the Competition Authority prosecutes the cartels at random, the duration of the cartel when caught will also indicate how long the cartels will last. Of course, since it seems probable that most cartels were legal until the 1980's, there is an effective truncation of the distribution of cartel duration. References [1] A. Hunnes, J. Møen and K. G. Salvanes. Wage Structure and Labour Mobility in Norway Work in progress. [2] E. J. Green and R. H. Porter, Noncooperative Collusion under Imperfect Price Information, Econometrica 52 (1984), [3] J. J. Rotemberg and G. Saloner, A Supergame-Theoretic Model of Price Wars during Booms, American Economic Review 76 (1986), [4] S. athey, K. Bagwell and C. Sanchirico, Collusion and Price Rigidity, Review of Economic Studies 71 (2004), [5] C. Fershtman and A. Pakes, A Dynamic Game with Collusion and Price wars, RAND Journal of Economics 31 (2000), [6] R. H. Porter, A Study of Cartel Stability: The Joint Executive Committee, , The Bell Journal of Economics 14 (1983),

10 10 Pr[Cartels in Norway] [7] G. Ellison, Theories of Cartel Stability and the Joint Executive Committee, The RAND Journal of Economics 25 (1994), [8] P. G. Bryant and E. W. Eckard, Price Fixing: The Probability of Getting Caught, Review of Economics and Statistics 73 (1991), [9] K. Eichenwald, The Informant, Broadway Books, 2000.